Discussion of article "Self-adapting algorithm (Part IV): Additional functionality and tests"

 

New article Self-adapting algorithm (Part IV): Additional functionality and tests has been published:

I continue filling the algorithm with the minimum necessary functionality and testing the results. The profitability is quite low but the articles demonstrate the model of the fully automated profitable trading on completely different instruments traded on fundamentally different markets.

In the previous article, I demonstrated how the algorithm generates a signal for opening a position and analyzes several scales simultaneously for defining the maximum trend scale. The basic operation algorithm was described. The price series chart does not consist of one scale. The trend can be present on several scales at the same time, while there can be flat on other scales. This feature should be used to make a profit.

Here, a trend section is a segment, on which the trend continuation probability exceeds 50%, while a flat segment is one, on which the trend reversal probability exceeds 50%. In other words, if the previous block was growing, then in the trend section, the new block will also grow with a probability higher than 50%. On a flat chart, a growing block is most probably followed by a falling block. I described the proposed definition in detail in the article "What is a trend and is the market structure based on trend or flat?"

trend-flat

Figure 1. Trend and flat on different scales

Figure 1 shows a clearly visible bearish trend on 32 blocks of 0.00061. The trend is almost absent on 32 blocks with the scale of 0.00131. In most cases, there are simultaneously the scales featuring both trend and flat.

Author: Maxim Romanov

[Deleted]  

The results were less than modest with this approach. And martingale is usually associated with high risk but also with profitability.

I have long ago noticed that going into abstractions like entropy, distributions and so on is not very suitable for the market. The complexity of abstract concepts grows, but the results do not.

 
Maxim Dmitrievsky:

The results were less than modest with this approach. And martingale is usually associated with high risk but also profitability.

I have long ago noticed that going into abstractions like entropy, distributions and so on is not very suitable for the market. The complexity of abstract concepts grows, but the results do not.

The results may seem modest. But somewhere there is a model that can show income on any market and large time intervals without tuning, maybe I missed something?
In general, the results improve after modification of various functions. Now I have already improved the results 2-3 times, but it was not included in the article.
[Deleted]  
Maxim Romanov:
The results may seem modest. But somewhere there is a model that can show income on any market and large time intervals without adjustment, maybe I missed something?
In general, the results improve after modification of various functions. Now I have already improved the results 2-3 times, but it was not included in the article.

If the income means beating a bank deposit, it is possible. But there the interest will be accrued constantly, and here with drawdowns 2-3 times more than profit. And with an unknown outcome at the end of the reporting period.

I am writing about what I saw on the screens

 
Maxim Dmitrievsky:

If by income you mean beating a bank deposit, it is possible. But there the interest will be accrued constantly, and here with drawdowns 2-3 times more than the profit. And with an unknown outcome at the end of the reporting period.

I plan to beat bank deposit significantly. Yes, now there are drawdowns, but they can be significantly reduced. In the last modification, I managed to reduce the drawdown on one instrument 10 times with the same parameters, relative to the version shown in the article.
Everything is realisable, you just need to gradually improve the algorithm. Computers were also large and inefficient, it's a matter of time and resources. And it is without customisation, that's what is important. If you want to go back in time, you can always optimise the parameters and show a nice picture. But I don't need pictures, there are enough of them here without me.
It's certainly not PNB) but I think it's not bad either).
[Deleted]  
Maxim Romanov:
I plan to beat bank deposits significantly. Yes, there are drawdowns now, but they can be significantly reduced. In the latest modification, I managed to reduce the drawdown on one instrument by 10 times with the same parameters, relative to the version shown in the article.
Everything is realisable, you just need to gradually improve the algorithm. Computers were also large and inefficient, it's a matter of time and resources. And it is without customisation, that's what is important. If you want to go back in time, you can always optimise the parameters and show a nice picture. But I don't need pictures, there are enough of them here without me.
It's certainly not PNB) but I think it's not bad either).

Just commented on what I saw, without any deep analysis :)

 
Maxim Dmitrievsky:

The results were less than modest with this approach. And martingale is usually associated with high risk, but also with profitability.

I have long ago noticed that going into abstractions like entropy, distributions and so on is not very suitable for the market. The complexity of abstract concepts grows, but the results do not.

I think so too.

for grid TSs one should look first of all at the equity line on the tester's charts, not at the balance, in the examples given in the article, imho, there is nothing to look at - unpredictable behaviour of TSs in general.

 
Maxim Dmitrievsky:

Just commented on what I saw, without any deep analysis :)

I understand the questions, it is not always clear what is being done and why, what goals are being pursued, so comments are needed. And a look from the outside helps to understand where mistakes can be made.

 
Igor Makanu:

I think so, too.

for grid TSs you should look first of all at the equity line on the tester's charts, not at the balance, in the examples given in the article, imho, there is nothing to look at - unpredictable behaviour of the TS at all

unpredictable, but for 18 years of testing on 56 trading instruments it shows plus without optimisations. What predictable behaviour will be then?

 
Maxim Romanov:

unpredictable, but for 18 years of testing on 56 trading instruments it shows plus without optimisations. What predictable behaviour would be then?

Predictably, the TS discussed in the article simply depend on the starting deposit.

SZY: there was a saying that if you sit on the river bank for a long time, you can see the body of a passing enemy...imho, the testing period of 18 years repeats this saying.

 
Igor Makanu:

predictably, the TSs discussed from the article simply depend on the starting deposit

SZY: there was a saying that if you sit on the river bank for a long time, you can see the body of a passing enemy...imho, the testing period of 18 years approximately repeats this saying.

I get it, there is no constructive criticism of the mechanics used because: "I don't understand anything, but I don't like it, the yield is low".